v2026.916.0 v6.4.1
ruvnet/RuVectorSep 16, 2026 → Sep 19, 20262 days between releases
What's New
Introduces `diagnosing-superpowers` skill and Native plan execution with breaking review behavior changes.
Breaking Changes
- `executing-plans` runs the whole plan before a single review checkpoint.
- `BASE_SHA` changed to `git merge-base origin/main HEAD`.
- Plans with identical basenames no longer share a workspace.
- `review-package` now rejects empty or non-descendant `BASE..HEAD` ranges.
- `AGENTS.md` is now the canonical contributor guide replacing symlinks.
New Features
- `diagnosing-superpowers` skill to analyze transcripts and generate debug bundles.
- Native (inline) execution for plan tasks with a final whole-branch review.
- Plan handoff offering Subagent-driven and Native options with cost-based recommendations.
- "Review Focus" section in saved plans to identify implied failure modes.
- Brainstorming that reflects user intent back for correction before scaffolding.
Improvements
- Cheaper, faster plan runs via Native execution model.
- Support for OpenCode 2.0, Muse, and Qwen Code harnesses.
- Improved brainstorming that clarifies user intent.
- Plans now include a 'Review Focus' section.
- Native execution is now a complete engine vs. a stub.
Upgrade Notes
Plan execution behavior changed: plans now run entirely before a single review checkpoint. Update git `BASE_SHA` logic to use `git merge-base origin/main HEAD`. Ensure plan basenames are unique to prevent workspace overwrites.
The Two Releases
v2026.916.0
Sep 16, 2026· 1 month ago
A major release introducing a centralized 'Connections' architecture for AI credentials, AgentMail, and native chat connectors, while removing legacy features like cheap model profiles.
Full release notesv6.4.1
Sep 19, 2026· 1 month ago
v6.4.1 is the first release with these changes, introducing a new debugging skill, rebuilding plan execution to Native mode, and adding support for OpenCode 2.0, Muse, and Qwen Code harnesses.
Full release notes